The process of bootstrapping the chroot from sources was
failing for several reasons when run from behind a firewall
with proxies. The llvm build was failing due to inability
to checkout sources through subversion using the
subversion.eclass wrapper (the "normal" way to do this in
the ebuild environment). This was because the user's
subversion configuration (including proxy settings) was not
inherited from $HOME/.subversion into the in-chroot sandbox
used by subversion.eclass.
This change creates symbolic links in the subversion.eclass
sandboxes for host and target builds in the chroot to fix
any build that uses the normal subversion.eclass for
checkouts. The operation is done at enter_chroot time so
that it applies to both ordinary builds and chroot creation
(via early_enter_chroot).
BUG=none
TEST='cros_sdk --replace --enter' behind proxied firewall
Change-Id: I0af2128866bb95799dc07c728c75cf3f2a0af7a3
Reviewed-on: https://gerrit.chromium.org/gerrit/34291
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: paul drews <paul.drews@intel.com>
Tested-by: paul drews <paul.drews@intel.com>
CL:33868 (7b6f377c58) introduced a
breakage in the "cros_sdk --replace --bootstrap" scenario.
The make_chroot.sh script invokes early_enter_chroot before
invoking init_setup. The chroot/etc/profiles.d directory is
created in init_setup, but the referenced change was
expecting to create a file in that directory in the context
of early_enter_chroot before the directory was created.
This led to a "no such file or directory" error when trying
to create the file.
This change does a "mkdir -p" of the referenced directory
before putting things in it in the context of
early_enter_chroot. The filename is also fixed to the name
expected elsewhere in the scripts.
BUG=none
TEST=cros_sdk --replace --bootstrap
Change-Id: I6ac0467117d7b0dd413695153469b367d56c256c
Reviewed-on: https://gerrit.chromium.org/gerrit/34958
Commit-Ready: Brian Harring <ferringb@chromium.org>
Reviewed-by: Brian Harring <ferringb@chromium.org>
Tested-by: Brian Harring <ferringb@chromium.org>
This is forced by cros_sdk; in conjunction w/ this,
drop --distfiles and mangle the chroot on during entrance
dropping a symlink in the old /var/cache/distfiles location
pointing to the new mounted cache_dir location.
Additionally, thread CHROMEOS_CACHEDIR down through the end.
Do this without relying on a version upgrade script- we can't
require they be run before entering, thus we exploit the fact
that cros_sdk explicitly forces a write lock to do the upgrade,
if we see the old form we know we can do the upgrade w/out
worrying about collisions.
CQ-DEPEND=CL:33871
BUG=chromium-os:34457
TEST=manual testing.
Change-Id: I6805266e3ec683f05d3ba615f9e8840642a28e48
Reviewed-on: https://gerrit.chromium.org/gerrit/33868
Commit-Ready: Brian Harring <ferringb@chromium.org>
Reviewed-by: Brian Harring <ferringb@chromium.org>
Tested-by: Brian Harring <ferringb@chromium.org>
enter_chroot.sh was not updating /etc/hosts from the out-of-chroot
environment. Make it do that.
BUG=None
TEST=locally
Change-Id: Ieaa337ae90dbc0700c42fa7e4b96faf12d3968cb
Reviewed-on: https://gerrit.chromium.org/gerrit/34226
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Ryan Cui <rcui@chromium.org>
Tested-by: Ryan Cui <rcui@chromium.org>
This change was coopted from http://codereview.chromium.org/5331009/,
originally written by hungte@. And the coopted commit message:
It would be helpful if we could share some directories inside/outside the
chroot (e.g. editor configuration or the default Downloads directory). This
CL reads .local_mounts (just like .default_boards) from the "src/scripts"
folder, and mounts the directories whenever you do cros_sdk.
For safety concern, and to prevent the developer from accidentally deleting
their mounted files, the mounts are made read-only.
.local_mounts has a very simple syntax:
mount_path
or source_path(outside chroot) destination_path(inside chroot)
or # comments.
Examples:
/usr/share/vim/google
/home/XXX/Downloads /outside
BUG=chromium-os:34561
TEST=Manually:
1. Create ~/trunk/src/scripts/.local_mounts with following content:
# comment here
/usr/share/vim/google # test
/home/XXX/Downloads /outside
2. cros_sdk
3. ls -l /usr/share/vim/google/ # ensure dir is mounted correctly
ls -l /outside/ # ensure dir is mounted correctly
4. exit
5. mount | grep chroot # ensure nothing is left
Change-Id: I6f3400a436a825e8cdfcb18b788afe96ebba6757
Reviewed-on: https://gerrit.chromium.org/gerrit/33585
Tested-by: Michael Krebs <mkrebs@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: Michael Krebs <mkrebs@chromium.org>
These are a new type of credential baked into chrome/chromium for
developers without internal copies of ChromeOS, and not building internal
versions of Chrome.
We automatically move .googleapikeys into the chroot each time.
We don't overwrite the destination, so that people can configure keys the
way they want. If they just don't want to be bothered, the best thing happens
the easiest way. Get Keys, put them in home. Keep working.
BUG=chromium-os:34438
TEST=local
Change-Id: I08e5970c6092f7b789aa5efef52db93841996d8f
Reviewed-on: https://gerrit.chromium.org/gerrit/33771
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: Peter Mayo <petermayo@chromium.org>
Tested-by: Peter Mayo <petermayo@chromium.org>
Since sudo changes $HOME to /root, sudoed invocations of gsutil/boto
won't find the necessary credentials. This solves the problem by
installing a symlink at /root/.boto to the correct credentials file,
similar to how it's done for /home/$USER/.boto.
BUG=None
TEST=/root/.boto symlink created upon entering the chroot
Change-Id: I541556f836fa5d0b9708e5604218058401563fb3
Reviewed-on: https://gerrit.chromium.org/gerrit/32430
Reviewed-by: David James <davidjames@chromium.org>
Reviewed-by: Ryan Cui <rcui@chromium.org>
Commit-Ready: Gilad Arnold <garnold@chromium.org>
Tested-by: Gilad Arnold <garnold@chromium.org>
Do this via ensuring that any common.sh invoker
of raw umount (say a root script) sees our umount
path.
Additionally, inject into default profiles our override,
and via an upgrade scriptlet.
This is round two; originally appeared as CL:32088, was
reverted due to:
https://uberchromegw.corp.google.com/i/chromiumos/builders/chromiumos%20sdk/builds/2314/steps/BuildBoard/logs/stdio
The fix however is just adding a single sudo mkdir. :/
BUG=chromium-os:23443
TEST=cros_sdk --replace --bootstrap
TEST=cros_sdk --replace
Change-Id: I0dc7522a9c623f40081d4f138cea0c2c45171fea
Reviewed-on: https://gerrit.chromium.org/gerrit/32365
Commit-Ready: Brian Harring <ferringb@chromium.org>
Tested-by: Brian Harring <ferringb@chromium.org>
Reviewed-by: Chris Sosa <sosa@chromium.org>
Do this via ensuring that any common.sh invoker
of raw umount (say a root script) sees our umount
path.
Additionally, inject into default profiles our override,
and via an upgrade scriptlet.
BUG=chromium-os:23443
TEST=manual validation, trybot.
Change-Id: Ie2514f6e8d2e10a19ab8d11c8056177bc1a2fb4d
Reviewed-on: https://gerrit.chromium.org/gerrit/32088
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: Brian Harring <ferringb@chromium.org>
Tested-by: Brian Harring <ferringb@chromium.org>
In particular, put the sudoers.d setup into one script (making
updates to it easier in the future if necessary), and
centralize the proxied vars into a const in common.sh.
Thanks to Kevin McCray/Josh Triplett/Alexander Kanevsky for
pointing out the missing proxy variables, and fixes/cleanup.
BUG=None
TEST=https_proxy=blah cros_sdk -- bash -c 'echo $https_proxy'
TEST=build_packages behind a proxy.
TEST=cros_sdk --replace && \
RSYNC_PROXY=blah cros_sdk -- bash -c 'echo $RSYNC_PROXY'
Change-Id: I3165882dfd9c8b52d25c2b26d7ff9242c84c91bd
Reviewed-on: https://gerrit.chromium.org/gerrit/31185
Tested-by: Brian Harring <ferringb@chromium.org>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: Josh Triplett <josh@joshtriplett.org>
The code has been in here long enough - most people should be
transitioned over who are using gerrit-source. We've also already
removed the chrome projects from the default manifest, and things look
good so far.
BUG=chromium-os:32963
TEST=remote trybot
Change-Id: Idd5e3a2ad77ea86c7316a9d50f5da1a5fdf01d8b
Reviewed-on: https://gerrit.chromium.org/gerrit/31161
Reviewed-by: Brian Harring <ferringb@chromium.org>
Commit-Ready: Ryan Cui <rcui@chromium.org>
Tested-by: Ryan Cui <rcui@chromium.org>
This is used to build toolchains with specific env variables
BUG=chromium-os:33240
TEST=trybot x86-generic-toolchain-minor
Change-Id: I2bbdd7d013a15c57c590a0d660a210e0ae2a6695
Reviewed-on: https://gerrit.chromium.org/gerrit/30645
Tested-by: Zdenek Behan <zbehan@chromium.org>
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Zdenek Behan <zbehan@chromium.org>
Don't run .bash_logout after invocation of the hook, which clears the
screen, sending unnecessary escape characters.
BUG=None
TEST=Ran locally.
Change-Id: I6c466040e7169d304b892b85be6a5b0d578e7714
Reviewed-on: https://gerrit.chromium.org/gerrit/29645
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Ryan Cui <rcui@chromium.org>
Tested-by: Ryan Cui <rcui@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
See the bug for details.
BUG=chromium-os:32963.
TEST=Locally, remote trybots.
Change-Id: I33f5c42b36f3e06139036c299c2fc2c2ff026411
Reviewed-on: https://gerrit.chromium.org/gerrit/28543
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Ryan Cui <rcui@chromium.org>
Tested-by: Ryan Cui <rcui@chromium.org>
Originally, I patched enter_chroot.sh to stop the gvfs daemons to work
around an issue where these daemons would prevent loop devices from being
unmounted. See https://bugzilla.gnome.org/show_bug.cgi?id=677648
Unfortunately, temporarily stopping gvfs daemons has a bad side effects:
other GUI applications that rely on these daemons responding start hanging.
This can be reproduced, for example, by starting 'gedit'.
To fix these hangs, I'm just reverting my patches to enter_chroot.sh and
restoring the scripts to where they were before.
This reverts the following patches:
1. Stop gvfs daemons earlier during enter_chroot.
This reverts commit 0079158f73.
2. Revert "Stop the gvfsd-trash daemon during enter_chroot."
This reverts commit 654a00bd61.
3. Revert "Stop the automounting daemon whenever we're inside the chroot."
This reverts commit fae0a59e8b.
4. Revert "Clean up update_bootloaders.sh to avoid sleeping."
This reverts commit 0103b59138.
BUG=chromium-os:23443
TEST=Trybot run.
Change-Id: Ie9ff222fe5fc7232fd1fc39af129cc18531118c6
Reviewed-on: https://gerrit.chromium.org/gerrit/26922
Reviewed-by: Chris Wolfe <cwolfe@chromium.org>
Reviewed-by: Jon Kliegman <kliegs@chromium.org>
Reviewed-by: Brian Harring <ferringb@chromium.org>
Commit-Ready: David James <davidjames@chromium.org>
Tested-by: David James <davidjames@chromium.org>
The killall commands for stopping gvfs weren't working
consistently for two reasons:
1) They ran too late, after it already picked up the
mounts in cros_sdk.
2) killall sometimes can only access the first 15 characters
of a process name, so we should only match on these characters.
BUG=chromium-os:23443
TEST=Verify gvfs is properly stopped when entering the chroot
on precise systems.
Change-Id: I16aff4b0d9ac101083b63e06e55d50869479a152
Reviewed-on: https://gerrit.chromium.org/gerrit/26369
Reviewed-by: Pawel Osciak <posciak@google.com>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: David James <davidjames@chromium.org>
Tested-by: David James <davidjames@chromium.org>
For fresh sdk builds, gcc won't update and automatically see the
ccache tree for us. So make sure the perms are sane when people
enter the chroot. This will also automatically fix perms if/when
people manually delete the ccache dir (which sometimes happens on
the buildbots when people try to free up space).
BUG=None
TEST=`rm -rf distfiles/ccache/; cros_sdk` and see ccache dir get setup nicely
Change-Id: I5bcc86ebf696549b142a7ceb312eb8ec4be5e2bf
Reviewed-on: https://gerrit.chromium.org/gerrit/26257
Reviewed-by: David James <davidjames@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
This changelist adds code in common.sh to support collecting command
statistics before calling upload_command_stats to upload those stats
to an appspot instance.
The presence of a file at ~/.disable_build_stats_upload will disable
all uploading of build command stats.
BUG=chromium-os:27355
TEST=`build_packages --board=x86-generic` with missing appspot instance
shows upload error but still goes through with build.
TEST=`build_packages --board=x86-generic` with real appspot instance
completes upload with all expected stats and does not affect build.
TEST=From outside chroot:
`touch ~/.disable_build_stats_upload`
`cros_sdk`
'build_packages --board=x86-generic`
Nothing uploaded due to .disable file
TEST=Verified that putting 'set -u' in my fake build_packages
caused an exit in print_time_elapsed, then fixed unbound variable
in print_time_elapsed, then rerun passed.
TEST=`cbuildbot -g <cl> --lkgm mario-paladin` passed
TEST=`cbuildbot -g <cl> --lkgm link-release` passed
Change-Id: Ieb714522cb32d7558b661e4ee1a197d1fce2c516
Reviewed-on: https://gerrit.chromium.org/gerrit/26084
Tested-by: Matt Tennant <mtennant@chromium.org>
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Matt Tennant <mtennant@chromium.org>
This reverts commit 615fe57ff0.
This CL broke mod_image_for_recovery.sh with the following failure:
common.sh: line 684: $2: unbound variable
This bug is easy to fix, but suggests the CL needs more testing, so
we're going to revert for now and re-submit the CL once it's been
verified to pass on release trybots.
Example failure:
http://chromegw.corp.google.com/i/chromeos/builders/x86-mario%20canary/builds/2214/steps/Archive/logs/stdio
BUG=chromium-os:27355
TEST=None, since this is reverting a change that broke the tree.
Change-Id: I61d182e3dcee267a8d9dea3b547fa6a75140d974
Reviewed-on: https://gerrit.chromium.org/gerrit/26077
Reviewed-by: Doug Anderson <dianders@chromium.org>
Tested-by: David James <davidjames@chromium.org>
This changelist adds code in common.sh to support collecting command
statistics before calling upload_command_stats to upload those stats
to an appspot instance.
The presence of a file at ~/.disable_build_stats_upload will disable
all uploading of build command stats.
BUG=chromium-os:27355
TEST=`build_packages --board=x86-generic` with missing appspot instance
shows upload error but still goes through with build.
TEST=`build_packages --board=x86-generic` with real appspot instance
completes upload with all expected stats and does not affect build.
TEST=From outside chroot:
`touch ~/.disable_build_stats_upload`
`cros_sdk`
'build_packages --board=x86-generic`
Nothing uploaded due to .disable file
Change-Id: Iac071d1cc55a44335fc7c846960c7ae45fc93ed8
Reviewed-on: https://gerrit.chromium.org/gerrit/19401
Tested-by: Matt Tennant <mtennant@chromium.org>
Reviewed-by: Matt Tennant <mtennant@chromium.org>
Commit-Ready: Matt Tennant <mtennant@chromium.org>
The gvfsd-trash daemon interferes with mounting and unmounting
images, so it should also be disabled for user chroots.
The gvfsd-trash daemon automagically creates 'trash' subdirectory inside
any directory you mount and holds on to it for a little while, preventing
you from being able to unmount your mount point.
BUG=chromium-os:23443
TEST=Verify that this fix fixes cros_make_image_bootable for daisy on
Ubuntu precise.
Change-Id: I8df4c999ed57d7025b63c971390448c93d404e83
Reviewed-on: https://gerrit.chromium.org/gerrit/25243
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Anush Elangovan <anush@google.com>
Tested-by: David James <davidjames@chromium.org>
If automounting is enabled while we're inside the chroot, simple mount /
unmount commands will fail, causing image building commands to fail flakily.
This allows for quick mount / unmount.
BUG=chromium-os:23443
TEST='cros_sdk true' works now and does not print warnings on systems that
previously had race conditions in this case.
Change-Id: I18d725324cc42ab7c527d3b52479b6a90bc4172c
Reviewed-on: https://gerrit.chromium.org/gerrit/25166
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: David James <davidjames@chromium.org>
Commit-Ready: David James <davidjames@chromium.org>
The only way to trigger this is if references were set up, but
never instantiated.
BUG=chromium-os:31682
TEST=manual permission checking.
Change-Id: Ibfc9dadf838f554cd10411753b5769117b1b1d42
Reviewed-on: https://gerrit.chromium.org/gerrit/24932
Reviewed-by: Brian Harring <ferringb@chromium.org>
Tested-by: Brian Harring <ferringb@chromium.org>
Commit-Ready: Brian Harring <ferringb@chromium.org>
When bootstrapping for the first time, files in chroot/etc/ might not
exist, so we can't run `find` on them. This manifests itself currently
by spitting out the warning on all initial sdk boots:
find: `.../chroot/etc/resolv.conf': No such file or directory
People can find this confusing and cause sheriffs to waste time on the
wrong thing, so rework the code to avoid this.
BUG=None
TEST=`cros_sdk --delete ; cros_sdk` no longer warns about resolv.conf
Change-Id: I83f892e325e63e682aeb370a9dfc33e284e059d2
Reviewed-on: https://gerrit.chromium.org/gerrit/22845
Reviewed-by: Brian Harring <ferringb@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Since /etc/mtab could be stale, use /proc/mounts instead.
BUG=None
TEST=`cros_sdk` in diff terminals still works
Change-Id: I526e5173581820c6983fe3702493a0349c1232c3
Reviewed-on: https://gerrit.chromium.org/gerrit/22860
Reviewed-by: David James <davidjames@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
The "function" keyword is superfluous, not in POSIX, is inconsistent
between bash files, and generally makes me angry. So convert every
instance to the form:
foo() {
BUG=None
TEST=`cbuildbot x86-generic-paladin` works
Change-Id: I97f5ca30a3edfef7222b1e08ac23917dc613b556
Reviewed-on: https://gerrit.chromium.org/gerrit/22467
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Currently, if set -e spots a nonzero exit we basically have
no real debug information- it just stops immediately without stating
where or why. This forces our scripts to be stupidly verbose so
we can track roughly where they were, thus when they fail we can
use that information to localize the rough exit point.
Instead we should be traping that set -e induced exit and
outputing necessary debug information to run it down. This includes
outputing the relevant stack trace, or at least what we can get of
it.
The 'die' function is now enhanced to automatically dump the trace
that lead to it. For most consumers this is desired- however for
commandline parsing induced dies ("--board is missing" for example),
the trace is noise. For those cases, a 'die_notrace' function was
added that retains the original non-backtrace behaviour.
Example output via instrumenting cros_generate_breakpad_symbols
w/ the failing command '/bin/false' (nonzero exit code).
Before:
./cros_generate_breakpad_symbols monkeys --board=x86-alex
<no output at all, just exit code 1>
With this CL:
./cros_generate_breakpad_symbols monkeys --board=x86-alex
ERROR : script called: ./cros_generate_breakpad_symbols 'monkeys' '--board=x86-alex'
ERROR : Backtrace: (most recent call is last)
ERROR : file cros_generate_breakpad_symbols, line 207, called: main 'monkeys' '--board=x86-alex'
ERROR : file cros_generate_breakpad_symbols, line 163, called: die_err_trap '/bin/false' '1'
ERROR :
ERROR : Command failed:
ERROR : Command '/bin/false' exited with nonzero code: 1
BUG=chromium-os:30598
TEST=inject a failing command into a script, verify the output.
TEST=inject a 'command not found', verify the output
TEST=cbuildbot x86-generic-full --remote
TEST=cbuildbot arm-tegra2-full --remote
TEST=cbuildbot chromiumos-sdk --remote
Change-Id: I517ffde4d1bb7e2310a74f5a6455b53ba2dea86c
Reviewed-on: https://gerrit.chromium.org/gerrit/17225
Reviewed-by: Brian Harring <ferringb@chromium.org>
Tested-by: Brian Harring <ferringb@chromium.org>
Commit-Ready: Brian Harring <ferringb@chromium.org>
The private chromeos-overlay has shared credentials for read-only
access to the archive of CrOS images. When entering the chroot check
whether these credentials are available, and install them to ~/.boto
within the chroot unless that file already exists.
BUG=None
TEST=Applied patch and entered fresh private and public chroots.
Manually ran 'gsutil config' to replace the credentials;
this works as long as there is no ~/.boto.bak already.
Re-entered the chroot with manually-configured credentials.
Ran 'gsutil ls' with BOTO_CONFIG set to a board-specific boto.
Change-Id: Ib62dc28f90bce692a833b5fc9d4c56981acc98d1
Reviewed-on: https://gerrit.chromium.org/gerrit/21480
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: Chris Wolfe <cwolfe@chromium.org>
Commit-Ready: Chris Wolfe <cwolfe@chromium.org>
The refresh-packages builder needs to start using auth tokens for logging into
Google Docs, rather than email/password each time, so this copies ~/.gdata_token
into the chroot in the same way as ~/gdata_cred.txt.
BUG=chromium-os:23819
TEST=Put a file at ~/.gdata_token outside chroot, then run:
`cros_sdk true ; cmp {chroot/home/$USER/,~/}.gdata_token`, which passes.
Change-Id: Ib688a01aa88e0fd0f0211236bb0354813ef1fbb3
Reviewed-on: https://gerrit.chromium.org/gerrit/15550
Commit-Ready: Matt Tennant <mtennant@chromium.org>
Reviewed-by: Matt Tennant <mtennant@chromium.org>
Tested-by: Matt Tennant <mtennant@chromium.org>
First, add a enable_strict_sudo helpers that scripts that are
sudo strict can invoke. This does a sanity check bailing immediately
if invoked from a non sudo-keep-alive context.
Second, update enter_chroot.sh and make_chroot.sh to be strict.
While this is strict, that's the point. The aim of this is to block
the previous sudo interactive mess for spreading, let alone reappearing
in scripts/code that has been cleansed.
BUG=chromium-os:18393
TEST=cros_sdk --replace; in the midst of it, do sudo -k.
cros_sdk should thus bail out w/ an appropriate error.
CQ-DEPEND=I01bb1466cf027401fa387af7fad15e42fd33aea4
Change-Id: I76c5b87a812cc78c30a2eb1a0c56b9e438f4a98f
Reviewed-on: https://gerrit.chromium.org/gerrit/15294
Commit-Ready: Brian Harring <ferringb@chromium.org>
Reviewed-by: Brian Harring <ferringb@chromium.org>
Tested-by: Brian Harring <ferringb@chromium.org>
Purpose of this is to allow us to avoid re-downloading everything
every time we rebuild the chroot.
This maintains two directories; host and target. Future enhancement
involves collapsing this into one- this requires some host work however,
and has some potential gotchas in doing so.
Meanwhile, we now store distfiles in repo/distfiles/{host,target},
and mount bind repo/distfiles into /var/cache/distfiles.
An upgrade script in turn optimistically tries to move the content
into the new location; if it can't complete the move, it wipes the
content and the user has to redownload it (acceptable, if annoying).
BUG=chromium-os:13115
TEST=cbuildbot x86-generic-full
Change-Id: Iea96429df0e1fdc4ac0860fbce0daabc90c4c2a3
Reviewed-on: https://gerrit.chromium.org/gerrit/15189
Reviewed-by: Brian Harring <ferringb@chromium.org>
Tested-by: Brian Harring <ferringb@chromium.org>
Commit-Ready: Brian Harring <ferringb@chromium.org>
Now that make_chroot.sh lives alongside enter_chroot, we're able
to invoke it directly and use internal options not exposed through
cros_sdk. Thus enter_chroot grows a --early_make_chroot flag.
This flag bypasses the normal sudo usage, and runs the command
as root. It is needed and used by make_chroot for when sudo
may not yet exist (--bootstrap via stage3 lacks sudo).
For where sudo exists and we need access to the source tree,
distfiles, or profile sourcing, enter_chroot is directly used.
For all other invocations we use a single sudo chroot call to
bypass the overhead of enter_chroot.
This is a complete fix to chromium-os:25697 while addressing
all refactoring requests that came up in I8a6c5a26.
Finally, convert all comments into complete sentences, upper
case the leading word, etc to keep reviewers happy.
BUG=chromium-os:25697
TEST=Within a trybot, cros_sdk --bootstrap --replace
TEST=cros_sdk --replace
TEST=cros_sdk --delete
TEST=cros_sdk
Change-Id: I270ee7fc325ef0bea74c61505d25cdbb49a9a333
Reviewed-on: https://gerrit.chromium.org/gerrit/15322
Commit-Ready: Brian Harring <ferringb@chromium.org>
Reviewed-by: Brian Harring <ferringb@chromium.org>
Tested-by: Brian Harring <ferringb@chromium.org>
This adds repo --references awareness (which maps down to git shared object
pools; see git clone -s); specifically, it sets up the appropriate indirection
bindings back to the actual shared source for use from within the chroot.
Note that the approach is a bit complex, but it's required complexity- via
this approach the references are accessible both from within, and without
the chroot.
BUG=chromium-os:19939
TEST=cbuildbot x86-generic-full --clobber --notests --nouprev
TEST=cbuildbot x86-generic-full --notests --nouprev
CQ-DEPEND=I986f17503dc154234ecadd90f6975d7164117cdb
Change-Id: I8498863010cefe0bf5b8f20350fa45a5f2a093d3
Reviewed-on: https://gerrit.chromium.org/gerrit/13467
Tested-by: Brian Harring <ferringb@chromium.org>
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Brian Harring <ferringb@chromium.org>
On long running systems, the ssh-agent dir might go stale. This should
not prevent entering the cros chroot though, so skip it if it's invalid.
BUG=None
TEST=`cros_sdk` enters with ssh-agent dir bind mounted
TEST=`rm -rf /tmp/ssh-*/; cros_sdk` enters with ssh-agent dir skipped
Change-Id: I923b031612c0d37a896437c7355cac6c448eef82
Reviewed-on: https://gerrit.chromium.org/gerrit/12758
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
The current instructions for golo access include additions to the
~/.ssh/config file that are specific to Google distros of ssh. The
filtering mechanism in enter_chroot needed an update to include
these new options to filter out.
BUG=chromium-os:23322
TEST=With a golo-configured .ssh/config, enter chroot and test ssh:
> cros_sdk
> git remote update
Fetching ...
...
Change-Id: Ida9fcdff37ff0eb24c811ad789f82136dca7252f
Reviewed-on: https://gerrit.chromium.org/gerrit/12037
Tested-by: Matt Tennant <mtennant@chromium.org>
Reviewed-by: David James <davidjames@chromium.org>
Reviewed-by: Zdenek Behan <zbehan@chromium.org>
Commit-Ready: Matt Tennant <mtennant@chromium.org>
Every invocation of `sudo` delays things, so merge all of the mounts
into a single `sudo` command when possible. This saves over 1 second
on initial execution (out of ~4 seconds total).
BUG=None
TEST=`cros_sdk --enter true` still mounts & unmounts properly
Change-Id: Ibc66507dc21250a81207d2f940645eaebe93c79c
Reviewed-on: https://gerrit.chromium.org/gerrit/10901
Reviewed-by: David James <davidjames@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
The current code that syncs the user's ssh files from ~/.ssh/ to the
chroot assumes the user has these files in the first place. For the
fresh user and/or install, these might not exist. So ignore errors
from attempting to copy them. It's not like we attempt to "recover"
anyways, and this sync is more of a "let's make things smoother" than
a "we need this to run properly".
BUG=chromium-os:15005
TEST=`mv ~/.ssh ~/.ssh.bak; mkdir ~/.ssh; cros_sdk` and see no more warnings
Change-Id: I9cac2ecad7c64c088fbdfe55377c0429b3b7ab03
Reviewed-on: https://gerrit.chromium.org/gerrit/10922
Reviewed-by: David James <davidjames@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Ubuntu 11.10, for whatever reason, moved /dev/shm to /run/shm. Since
/run is not mounted in the chroot, /dev/shm is a dangling symlink and
various things that require shared memory start failing.
If /run and /run/shm exist, bind-mount them in the chroot.
Because /run doesn't yet exist in the SDK tarballs, try to create it
(and /run/shm if necessary) as root if it and can't be created as a
normal user.
BUG=chromium-os:19871
TEST=Run cros_sdk, verify that /dev/shm has the expected contents
Change-Id: I61583c1c0d409c1234fa8d8930a9b64544c9a8e7
Reviewed-on: https://gerrit.chromium.org/gerrit/10222
Tested-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: David James <davidjames@chromium.org>
The sync pid file is created without using sudo which means we should be
able to delete it without using sudo. By default, run `rm` directly and
if it fails, fall back to sudo like we historically have.
BUG=None
TEST=`cros_sdk --enter true`; see pid file deleted
Change-Id: I26d898f6d594eb9ea4652335468345dd11303122
Reviewed-on: http://gerrit.chromium.org/gerrit/8644
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
We call `readlink -f` on the chroot mount point to resolve any symlinks
in the path (since that is what `mount` records). But the only paths
that can have symlinks are in the base chroot mount. Everything below
that we know are not going to be symlinks (since we've set up the paths
ourselves). So process the chroot mount point with readlink once and
reuse that value everywhere else.
BUG=None
TEST=`cros_sdk --enter true` still mounts & unmounts properly
Change-Id: Id1e734d20c0cb766f5490583b793930af77b3b14
Reviewed-on: http://gerrit.chromium.org/gerrit/8645
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
The chrome root code doesn't attempt to mount the tree if there is no
source root configured. So re-order the code from:
- is chrome source mounted ?
- does a chrome source exist ?
- mount chrome source
to the more logical:
- does a chrome source exist ?
- use ensure_mounted to check+mount
This lets us use the mount cache for the chrome mounting and unifies
duplicated mount/checking logic.
BUG=None
TEST=`cros_sdk --enter` still works with simultaneous runs
TEST=`cros_sdk --enter -- ls` still works
Change-Id: I7e6af9dd7f65cefa04438c2862c931f06237060a
Reviewed-on: http://gerrit.chromium.org/gerrit/8032
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
The depot_tools mount code is exactly like ensure_mounted except that it
prints a warning instead of dying when things can't be mounted. So take
the current api and extend it slightly to support warning or dying. This
also lets us re-use the existing mount cache and avoid the forks as well
as clean up duplicated code.
BUG=None
TEST=`cros_sdk --enter` still works with simultaneous runs
TEST=`cros_sdk --enter -- ls` still works
Change-Id: I89336778b6aa16191e79d900a51774929cadf06b
Reviewed-on: http://gerrit.chromium.org/gerrit/8031
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
The cros_sdk tool runs a lot of helper programs which can be replaced
with bash builtins, or condensed multiple calls into a single one. By
themselves they aren't that slow, but add them all up and run them a
whole lot, and it starts to make a difference.
External programs to bash internals:
- dirname $f -> ${f%/*}
- f=$(cat $f) -> f=$(<$f)
- which f -> type -P f
Simpler expressions:
- [[ ( ... ) ]] -> [[ ... ]]
- eval v=\$$f -> v=${!f}
Common/clearer expressions:
- ! var=$(cmd) -> var=$(cmd || :)
Condensed tools:
- sort | uniq -> uniq -u
BUG=None
TEST=`cros_sdk --enter` still works with simultaneous runs
TEST=`cros_sdk --enter -- ls` still works
Change-Id: Ice5e5e252237082a2249990644e051895d61d1fd
Reviewed-on: http://gerrit.chromium.org/gerrit/8029
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Most of the time the files in the chroot /etc/ are already changed to
the proper owner. Only on the first run do they need to be changed.
So check the owner before doing the sudo as that is a bit faster. If
we do end up needing to run sudo, bundle the commands up to avoid having
to execute sudo multiple times.
BUG=None
TEST=`sudo chown root chroot/etc/resolv.conf && cros_sdk --enter`; see /etc/resolv.conf owned by me
Change-Id: Ifd974ace168ab309fdc2e7583d135ee23576a5bb
Reviewed-on: http://gerrit.chromium.org/gerrit/8418
Reviewed-by: David James <davidjames@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Since we grab a lock before mounting anything, we know that the relevant
list of mounts won't change (since we haven't changed it yet). So we can
cache the output of `mount` and re-use that in every subsequent check.
BUG=None
TEST=`cros_sdk --enter` still works with simultaneous runs
TEST=`cros_sdk --enter -- ls` still works
Change-Id: I27c98c923761fb777686632fe6e6604ca543cfb6
Reviewed-on: http://gerrit.chromium.org/gerrit/8030
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Checking the permission of files is faster than assuming they're broken
and then running `sudo chmod`, so do just that.
BUG=None
TEST=run `cros_sdk --enter` with bad perms and see them fixed
TEST=run `cros_sdk --enter` with correct perms and see sudo skipped
TEST=the buildbot failure was due to other changes, not this one
Change-Id: Ie69b4e766e2e3652944e4723d091ce589d07a4f6
Reviewed-on: http://gerrit.chromium.org/gerrit/8028
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
No need to call `git var | sed` multiple times when we can call it once
and then use bash string parsing routines to extract the relevant info.
BUG=None
TEST=delete .gitconfig in chroot; enter chroot; verify .gitconfig is up-to-date
Change-Id: I900e4241a5c53bc46c90eb0f2a01152f4bfba577
Reviewed-on: http://gerrit.chromium.org/gerrit/8027
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
First let's add a chroot upgrade hook that'll declare a small list of
locales to speed up glibc upgrades.
The other blocker was that `cros_sdk --enter` took an inordinate amount
of time to execute (`locale-gen -u` specifically) when the number of
existing locales is huge. This seems to be a bug in the bash/glibc
stack which is resolved in newer versions, but we can workaround the
issue by forcing locale-gen to run in a C locale. The tool itself does
not care about its locale and we silence its output.
BUG=chromium-os:20378
TEST=`cros_sdk --enter`; see no locales in /etc/locale.gen; run `./build_packages` and see upgrade hook work
TEST=generate all 400 locales; `cros_sdk --enter`; see that it was quick
Change-Id: I8fcc5e26bd8e1bcfd52b6a6c7ef3cacf0a252081
Reviewed-on: http://gerrit.chromium.org/gerrit/7806
Reviewed-by: David James <davidjames@chromium.org>
Reviewed-by: Zdenek Behan <zbehan@chromium.org>
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
Running `sudo` at all is a bit poky, so let's avoid modprobing fuse when
we know it already exists because it's listed in /proc/filesystems.
The relative speed:
time sudo modprobe fuse -> 0.164s
time grep -q fuse /proc/filesystems -> 0.002s
BUG=None
TEST=enable `set -x`; run `sudo modprobe fuse`; run `cros_sdk true`; see modprobe skipped
Change-Id: I07213e7b0607ca65df2a8a3ae22eb4a49c8bb888
Reviewed-on: http://gerrit.chromium.org/gerrit/7824
Commit-Ready: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Tested-by: Mike Frysinger <vapier@chromium.org>
